Implementation and Analysis of Fuzzy Inference System (FIS) and Adaptive Neuro-Fuzzy Inference System (ANFIS) for Irrigation

Abstract

Efficient water management in agriculture is crucial due to dynamic environmental conditions and increasing resource scarcity. Fuzzy Inference System (FIS) is widely applied in irrigation control for its ability to handle uncertaintys using rule-based domain knowledge. However, conventional FIS lacks adaptability to environmental changes, limiting its long-term accuracy and responsiveness. Adaptive Neuro-Fuzzy Inference System (ANFIS) addresses this limitation by combining fuzzy logic with neural network learning, enabling automatic adjustment of model parameters based on data patterns. This study compares the performance of FIS and ANFIS in predicting optimal irrigation levels based on soil moisture, air temperature, relative humidity, and solar radiation. A synthetic dataset of 1,000 samples simulating realistic agricultural conditions was generated and normalized to improve computational consistency. The FIS model uses triangular membership functions and five expert-defined fuzzy rules, while ANFIS employs Gaussian membership functions with parameters optimized using the ADAM algorithm over 50 training epochs. Results show that ANFIS outperforms FIS, lowering RMSE from 0.13 to 0.07, halving MAE from 0.10 to 0.05, and increasing R² from 0.85 to 0.93, indicating a substantially better predictive performance. This study demonstrates that ANFIS is more adaptive, accurate, and computationally efficient, contributing to the advancement of intelligent and sustainable irrigation systems in precision agriculture.